-
Institutional Investors and Equity Returns:Are Short-term
Institutions Better Informed?
Xuemin (Sterling) YanUniversity of Missouri - Columbia
Zhe ZhangSingapore Management University
We show that the positive relation between institutional
ownership and futurestock returns documented in Gompers and Metrick
(2001) is driven by short-term institutions. Furthermore,
short-term institutions trading forecasts future stockreturns. This
predictability does not reverse in the long run and is stronger for
smalland growth stocks. Short-term institutions trading is also
positively related to futureearnings surprises. By contrast,
long-term institutions trading does not forecast futurereturns, nor
is it related to future earnings news. Our results are consistent
with theview that short-term institutions are better informed and
they trade actively to exploittheir informational advantage. (JEL
G12, G14, G20)
This article examines the relation between institutions
investment horizonsand their informational roles in the stock
market. Although a large body ofliterature has studied the behavior
of institutional trading and its impact onasset prices and
returns,1 the informational role of institutional investorsremains
an open question. Gompers and Metrick (2001) document apositive
relation between institutional ownership and future stock
returns.However, they attribute this relation to temporal demand
shocks ratherthan institutions informational advantage. Nofsinger
and Sias (1999)find that changes in institutional ownership
forecast next years returns,suggesting that institutional trading
contains information about futurereturns. In contrast, Cai and
Zheng (2004) find that institutional trading
We thank Paul Brockman, Kalok Chan, Charles Cao, Bhagwan
Chowdhry, Dan French, AllaudeenHameed, Grace Hao, John Howe, Inder
Khurana, Tina Martin, Sandra Mortal, Shawn Ni, AndyPuckett,
Avanidhar Subrahmanyam, Marti Subrahmanyam, and seminar
participants at the SingaporeManagement University and University
of Missouri Columbia for helpful comments. We are grateful forthe
comments of an anonymous referee and Terrance Odean (the editor).
We thank I/B/E/S for providinganalyst earnings forecast data. Zhang
acknowledges financial support from the Research Committeeof
Singapore Management University and the Lee Foundation. Address
correspondence to Zhe Zhang,Singapore Management University, Lee
Kong Chian School of Business, 50 Stamford Road, 178899,Singapore,
or e-mail: [email protected].
1 For institutional preferences, see, for example, Del Guercio
(1996), Falkenstein (1996), Gompers andMetrick (2001), and Bennett,
Sias, and Starks (2003). For institutional trading patterns such as
herding andmomentum trading and its impact on stock returns, see,
for example, Lakonishok, Shliefer, and Vishny(1992), Grinblatt,
Titman, and Wermers (1995), Nofsinger and Sias (1999), Wermers
(1999), Badrinathand Wahal (2001), Bennett, Sias, and Starks
(2003), Griffin, Harris, and Topaloglu (2003), Sias (2003),Cai and
Zheng (2004), Sias (2004), Sias (2005), and Sias, Starks, and
Titman (2005).
The Author 2007. Published by Oxford University Press on behalf
of The Society for Financial Studies.All rights reserved. For
Permissions, please email:
[email protected]:10.1093/rfs/hhl046
Advance Access publication January 3, 2007
-
has negative predictive ability for next quarters returns.
Bennett, Sias,and Starks (2003) show that the evidence of
institutions ability to forecastreturns is sensitive to how
institutional trading is measured.
One potential reason for the mixed results regarding
institutionalinvestors informational role is that most studies in
this literature focuson all institutional investors as a group.
While institutional investorsshare some important commonalities,
they are far from homogeneous.An important dimension of
heterogeneity is the investment horizon.Institutions may have
different investment horizons because of differencesin investment
objectives and styles, legal restrictions, and
competitivepressures; in addition, their investment horizons may
differ because oftheir different informational roles.
There are several reasons why one might expect institutions
withdifferent investment horizons to be differentially informed.
First, if someinstitutional investors possess superior information
and can regularlyidentify undervalued or overvalued stocks, we
would expect theseinstitutions to trade frequently to exploit their
informational advantageor skill [e.g., Grinblatt and Titman (1989)
and Wermers (2000)]. On theother hand, institutional investors
possessing limited information wouldtrade more cautiously.
Therefore, institutions that trade more actively(short-term
institutions) would be better informed than those that tradeless
actively (long-term institutions).2 Second, one might argue that
long-term institutions trade infrequently because they trade only
on the basis ofinformation. On the other hand, short-term
institutions might also trade onthe basis of noise, perhaps owing
to overconfidence [e.g., Odean (1998) andBarber and Odean (2000)].
In this case, it would appear on average thatlong-term institutions
are better informed than short-term institutions.Third, it is also
possible that both short- and long-term institutions areinformed.
However, short-term institutions are better at collecting
andprocessing short-term information, while long-term institutions
are betterat collecting and processing long-term information. As a
result, short-terminstitutions would be better informed in the
short run while long-terminstitutions would be better informed in
the long run.
The purpose of this article is to empirically examine the
informationalroles of short- and long-term institutions.
Specifically, using quarterlyinstitutional holdings for the period
from 1980 to 2003, we constructan investment horizon measure based
on institutions portfolio turnover,which is similar to that of
Gaspar, Massa, and Matos (2005). We thenclassify institutions into
short- and long-term based on this measure. Inour empirical
analyses, we first examine whether short- and long-term
2 Recent accounting literature [Ke and Petroni (2004)] provides
evidence that transient institutional investorscan predict a break
in a string of consecutive quarterly earnings increases, suggesting
that short-terminstitutions are informed.
894
The Review of Financial Studies / v 22 n 2 2009
-
Institutional Investors and Equity Returns
institutions have different preferences for stock
characteristics. We thenexamine the extent to which the investment
horizon affects the relationbetween institutional ownership and
future stock returns. More impor-tantly, we investigate whether
short- and long-term institutional tradingcontains information
about future stock returns and future earnings.
We find that both short- and long-term institutions prefer
larger stocksand stocks with higher book-to-market ratios, share
price, and volatility.Compared to long-term institutions,
short-term institutions prefer youngerfirms, and firms with higher
turnover and lower dividend yield. In addition,we find that only
short-term institutions are momentum traders.
Consistent with Gompers and Metrick (2001), we find a
significant andpositive relation between total institutional
ownership and one-quarter-ahead and one-year-ahead stock returns.
Moreover, we show that thispositive relation is almost entirely
driven by short-term institutions: Thepredictive power of total
institutional ownership is completely subsumedby short-term
institutional ownership. In contrast, long-term
institutionalownership does not have incremental predictive power
for future returns.
To test whether the predictive ability of institutional
ownership is dueto temporal demand shocks or institutions
informational advantage, wedecompose the current institutional
ownership into lagged institutionalownership and institutional
trading.3 Our results show that laggedinstitutional ownership by
short-term institutions forecasts future returns,suggesting that
demand shocks have an impact on returns. Moreimportantly,
short-term institutions trading also strongly predicts
futurereturns. Since our results are obtained after controlling for
variousstock characteristics including size, book-to-market, and
past returns,they cannot be explained by short-term institutions
following certaininvestment styles that have been shown to explain
cross-sectional stockreturns. In particular, our results are not
driven by the momentum effect.The predictive power of short-term
institutional trading is thus consistentwith the hypothesis that
short-term institutions are informed.
In contrast to the results for short-term institutions, we find
no evidencethat either the level or the change in long-term
institutional ownership issignificantly related to future stock
returns. This result is not driven byinstitutions that follow index
investment strategies. Long-term institutionsdo not predict future
returns after we exclude those fund families thatspecialize in
index funds, or during the first half of our sample
period(19801991) when index products are relatively
undeveloped.
To the extent that short-term institutions have an
informationaladvantage, we would expect their advantage to be
greater for small
3 Gompers and Metrick (2001) argue that since the institutional
holdings are fairly stable over time, thelagged institutional
holdings should be almost as good a proxy for temporal demand
shocks as currentinstitutional ownership. At the same time, changes
in the institutional holdings are a more precise measureof
informational advantage. See Gompers and Metrick (2001) for a
detailed discussion.
895
-
and growth stocks, which tend to have greater information
uncertaintyand are more difficult to value. Consistent with this
prediction, we find thatshort-term institutional trading has
stronger predictive power for smalland growth stocks than for large
and value stocks.
Our main results also hold with a portfolio approach. A
zero-investmentstrategy that is long in the portfolio of stocks
with the largest increasein short-term institutional holdings and
short in the portfolio of stockswith the largest decrease in
short-term institutional holdings generates2.16% [1.62% after
adjusting for Daniel, Grinblatt, Titman, and Wermers(1997)
benchmark returns] over the first year after portfolio formation.
Bycontrast, there is no significant return difference among
portfolios sortedby changes in long-term institutional
ownership.
We also examine whether institutional trading is related to
futureearnings news. We find that stocks that experience the
largest increasein short-term institutional holdings have
significantly higher earningssurprises and earnings announcement
abnormal returns over thesubsequent four quarters than stocks that
experience the largest decreasein short-term institutional
holdings. In contrast, we find little evidencethat trading by
long-term institutions is related to either future
earningssurprises or earnings announcement abnormal returns. These
resultsprovide evidence that short-term institutions possess
superior informationabout future earnings.
An alternative explanation for our results is that short-term
institutionalinvestors pressure managers to maximize short-run
profits at the expenseof long-run firm value [the short-term
pressure hypothesis; see forexample, Porter (1992), Bushee (1998,
2001)]. In particular, Bushee (1998)finds evidence that firms with
higher transient institutional ownership aremore likely to
underinvest in long-term, intangible projects such as R&Dto
reverse an earnings decline. The short-term pressure hypothesis
mighthelp explain the short-run predictive ability of short-term
institutionalownership and trading, but it also predicts long-run
price reversal forstocks held or traded by short-term institutions.
To test this prediction,we examine the relation between
institutional ownership and trading andfuture stock returns up to
three years. We find no evidence of long-runprice reversal for
stocks held or recently traded by short-term
institutionalinvestors, suggesting that our results cannot be
explained by the short-term pressure hypothesis. We also find no
evidence that either the holdingsor trading by long-term
institutional investors predicts long-run stockreturns. This result
is inconsistent with the hypothesis that long-terminstitutions are
better informed about long-run returns.4
4 Consistent with our finding, Ke and Ramalingegowda (2005)
provide evidence that long-term institutionalinvestors do not
possess private information about long-run earnings growth.
896
The Review of Financial Studies / v 22 n 2 2009
-
Institutional Investors and Equity Returns
To our knowledge, this is the first article in the finance
literature thatfocuses on the informational roles of short- and
long-term institutions.5
In particular, our results suggest that short-term institutions
are betterinformed, and that they trade actively to exploit their
informationaladvantage. Our results do not imply market
inefficiency: rather, they areconsistent with the idea that the
stock market is informationally efficient(Grossman and Stiglitz
(1980)) because short-term institutions mustexpend resources in
collecting, processing, and trading on information.
The rest of the article is organized as follows. Section 1
describes thedata and presents descriptive statistics. Section 2
examines institutionalpreferences. Section 3 investigates the
impact of institutional holdings andtrading on future stock returns
for both short- and long-term institutionalinvestors. Section 4
concludes.
1. Data, Variables, and Descriptive Statistics
1.1 Data and sampleThe data for this study come from four
sources. We obtain quarterlyinstitutional holdings for all common
stocks traded on NYSE, AMEX,and NASDAQ for the period from the
fourth quarter of 1979 to the fourthquarter of 2003 from Thomson
Financial. The Securities and ExchangesCommission (SEC) requires
that all investment managers with discretionover 13F securities
worth $100 million or more report all equity positionsgreater than
10,000 shares or $200,000 to the SEC at the end of eachquarter.
Institutional ownership (hereafter IO) for each stock is defined
asthe number of shares held by institutional investors divided by
the totalnumber of shares outstanding. We exclude those
observations with totalinstitutional ownership greater than
100%.
We obtain stock return, share price, number of shares
outstanding, andturnover from the Center for Research in Security
Prices (CRSP) monthlytapes for all NYSE/AMEX/NASDAQ stocks. We
obtain book value ofequity, cash dividend, and quarterly earnings
announcement dates fromCOMPUSTAT. Finally, analysts earnings
forecasts are from I/B/E/S.
1.2 Classification of short- and long-term institutionsWe
classify institutional investors into short- and long- term
investors onthe basis of their portfolio turnover over the past
four quarters. Specifically,each quarter, we first calculate the
aggregate purchase and sale for each
5 Several articles examine the relation between portfolio
turnover and mutual fund performance. Grinblattand Titman (1989)
find a positive relation between turnover and pre-expense portfolio
performance, whileCarhart (1997) finds that fund turnover is
negatively related to net fund returns. Chen, Jegadeesh, andWermers
(2000) show that, unlike in our study, both high- and low-turnover
mutual funds have stockpicking skillsthe stocks they buy outperform
those they sell. Moreover, they find little difference inskills
between high and low turnover funds based on their trading.
897
-
institution:
CR buyk,t =Nk
i = 1Sk,i,t > Sk,i,t1
|Sk,i,tPi,t Sk,i,t1Pi,t1 Sk,i,t1Pi,t | (1)
CR sellk,t =Nk
i = 1Sk,i,t Sk,i,t1
|Sk,i,tPi,t Sk,i,t1Pi,t1 Sk,i,t1Pi,t | (2)
where Pi,t1 and Pi,t are the share prices for stock i at the end
of quartert 1 and t , and Sk,i,t1 and Sk,i,t are the number of
shares of stock i held byinvestor k at the end of quarter t 1 and t
, respectively. We adjust stocksplits and stock dividends by using
the CRSP price adjustment factor.CR buyk,t and CR sellk,t are
institution ks aggregate purchase and salefor quarter t ,
respectively. Institution ks churn rate for quarter t is
thendefined as:
CRk,t min (CR buyk,t , CR sellk,t )Nki=1
Sk,i,tPi,t + Sk,i,t1Pi,t12
(3)
The above definition is similar in spirit to Gaspar, Massa, and
Matos(2005). The main difference is that we use the minimum of
aggregate pur-chase and sale, whereas Gaspar, Massa, and Matos
(2005) use the sum ofaggregate purchase and sale. The advantage of
our measure is that it mini-mizes the impact of investor cash flows
on portfolio turnover.6 CRSP uses avery similar approach to
calculate mutual fund turnover. Next, we calculateeach institutions
average churn rate over the past four quarters as:
AVG CRk,t = 143
j=0CRk,tj (4)
Given the average churn rate measure, each quarter we sort all
institutionalinvestors into three tertile portfolios based on AVG
CRk,t . Those rankedin the top tertile (with the highest AVG CRk,t
) are classified as short-terminstitutional investors and those
ranked in the bottom tertile are classifiedas long-term
institutional investors. Finally, for each stock, we define
theshort-term (long-term) institutional ownership (hereafter SIO
and LIO)as the ratio between the number of shares held by
short-term (long-term)institutional investors and the total number
of shares outstanding.
6 Alexander, Cici, and Gibson (2006) show that investor
flow-induced trading contains little information.
898
The Review of Financial Studies / v 22 n 2 2009
-
Institutional Investors and Equity Returns
1.3 Firm characteristicsLike Gompers and Metrick (2001), we
focus our analysis on the followingten firm characteristics:
MKTCAPmarket capitalization calculated as share price timestotal
shares outstanding using data from CRSP.
AGEfirm age calculated as the number of months since first
returnappears in CRSP.
DPdividend yield calculated as cash dividend divided by
shareprice.
BMbook-to-market ratio, book value for the fiscal year
endedbefore the most recent June 30, divided by market
capitalization ofDecember 31 during that fiscal year.
PRCshare price from CRSP. TURNaverage monthly turnover over the
past 3 months. VOLvolatility estimated as the standard deviation of
monthly
returns over the previous two years. SP500dummy variable for
S&P 500 index membership. RETt3,t cumulative gross return over
the past three months. RETt12,t3 cumulative gross return over the
nine months
preceding the beginning of filing quarter.
Following Gompers and Metrick, we use natural log for all the
abovevariables except for the S&P500, RETt3,t , and
RETt12,t3.
1.4 Descriptive statisticsWe compute, for each quarter, mean
cross-sectional institutionalownership and firm characteristics for
the period from the third quarter of1980 to the fourth quarter of
2003. Panel A of Table 1 reports the time-series mean, median,
maximum, minimum, and standard deviation of these94 cross-sectional
averages. The average institutional ownership is 25.1%over our
sample period. This result is similar to Bennett, Sias, and
Starks(2003), who report a 23% average institutional ownership for
the periodfrom 1983 to 1997. On average, short-term institutions
hold 7.91% of totalshares outstanding, while long-term institutions
hold 6.56% of all shares.
The average firm has a market capitalization of $961.44 million,
adividend yield of 2.21%, a book-to-market ratio of 0.74, and
approximately12 years of CRSP return data. The monthly volatility
and turnover forthe average firm are 13.59% and 7.8% respectively.
The average numberof stocks in our sample is 5911. In comparison,
Bennett, Sias, and Starks(2003) report an average of 5425 stocks in
their sample for 19831997.
Panel B of Table 1 reports the time-series average of the
cross-sectional correlations between institutional ownership and
various firmcharacteristics. Total IO is positively correlated with
size, age, price,turnover, and S&P 500 dummy, while negatively
correlated with volatility.
899
-
Tab
le1
Descriptive
statistics
Pan
elA
:tim
e-se
ries
stat
isti
csof
cros
s-se
ctio
nala
vera
ges
Stan
dard
Mea
nM
edia
nM
axim
umM
inim
umde
viat
ion
Tot
alin
stit
utio
nalO
wne
rshi
p
IO(%
)25
.10
24.8
538
.73
15.7
16.
13Sh
ort-
term
inst
itut
iona
lOw
ners
hip
SI
O(%
)7.
918.
1211
.46
5.47
1.58
Lon
g-te
rmin
stit
utio
nalO
wne
rshi
p
LIO
(%)
6.56
6.78
9.85
2.95
1.71
Mar
ket
capi
taliz
atio
n
MK
TC
AP
($m
illio
n)96
1.44
682.
5924
68.6
031
2.88
659.
54A
ge(m
onth
s)14
7.09
145.
0417
1.08
131.
1411
.34
Div
iden
dyi
eld
D
P(%
)2.
212.
103.
481.
650.
46B
ook-
to-m
arke
tB
M0.
740.
691.
350.
430.
20P
rice
P
RC
($)
21.3
319
.26
35.9
013
.20
5.21
Tur
nove
rT
UR
N(%
)7.
807.
2718
.14
3.01
3.11
Vol
atili
ty
VO
L(%
)13
.59
12.8
818
.78
11.1
82.
01L
agge
dth
ree-
mon
thre
turn
R
ET
t3,
t(%
)4.
193.
5532
.22
29.
1711
.38
Lag
ged
nine
-mon
thre
turn
R
ET
t12
,t3
(%)
13.5
812
.96
93.9
02
5.68
20.9
6N
umbe
rof
stoc
ks59
115,
941
7759
3453
1,21
2
Pan
elB
:tim
e-se
ries
mea
nof
cros
s-se
ctio
nalc
orre
lati
ons
betw
een
inst
itut
iona
low
ners
hip
and
stoc
kch
arac
teri
stic
s
MK
TC
AP
AG
ED
PB
MP
RC
TU
RN
VO
LSP
500
RE
Tt
3,t
RE
Tt
12,t
3
IO0.
200.
300.
120.
090.
130.
190
.24
0.39
0.02
0.06
SIO
0.09
0.10
0.01
0.00
0.09
0.31
0.0
80.
200.
040.
12L
IO0.
220.
380.
150.
120.
100.
010
.23
0.40
0.00
0.00
900
The Review of Financial Studies / v 22 n 2 2009
-
Institutional Investors and Equity Returns
Pan
elC
:tim
e-se
ries
mea
nof
cros
s-se
ctio
nalc
orre
lati
ons
betw
een
stoc
kch
arac
teri
stic
s
MK
TC
AP
AG
ED
PB
MP
RC
TU
RN
VO
LSP
500
RE
Tt
3,0
RE
Tt
12,t
3
MK
TC
AP
1.00
AG
E0.
311.
00D
P0.
030.
101.
00B
M0
.02
0.05
0.68
1.00
PR
C0.
140.
090.
030.
001.
00T
UR
N0
.01
0.0
90.
000.
040.
011.
00V
OL
0.1
20
.24
0.2
00
.02
0.1
00.
241.
00SP
500
0.44
0.50
0.04
0.0
20.
100.
030
.17
1.00
RE
Tt
3,t
0.02
0.01
0.0
20
.06
0.04
0.10
0.07
0.00
1.00
RE
Tt
12,t
30.
030.
010
.03
0.0
90.
050.
140.
080.
000.
021.
00
Thi
sta
ble
repo
rts
the
desc
ript
ive
stat
isti
cs.
The
sam
ple
peri
odis
from
1980
:Q3
to20
03:Q
4.In
stit
utio
nal
hold
ings
are
obta
ined
from
Tho
mso
nF
inan
cial
.Sto
ckch
arac
teri
stic
sar
efr
omth
eC
RSP
and
CO
MP
UST
AT
data
base
.IO
isto
tali
nsti
tuti
onal
owne
rshi
p.SI
Ois
shor
t-te
rmin
stit
utio
nal
owne
rshi
p.L
IOis
long
-ter
min
stit
utio
nalo
wne
rshi
p.A
nin
stit
utio
nali
nves
tor
iscl
assi
fied
asa
shor
t-te
rmin
vest
orif
its
past
4-qu
arte
rtu
rnov
erra
tera
nks
inth
eto
pte
rtile
.A
nin
stit
utio
nal
inve
stor
iscl
assi
fied
asa
long
-ter
min
vest
orif
its
past
4-qu
arte
rtu
rnov
erra
tera
nks
inth
ebo
ttom
tert
ile.
MK
TC
AP
ism
arke
tca
pita
lizat
ion.
Age
isfir
mag
em
easu
red
asnu
mbe
rof
mon
ths
sinc
efir
stre
turn
appe
ars
inth
eC
RSP
data
base
.DP
isdi
vide
ndyi
eld.
DP
isw
inso
rize
dat
the
99th
perc
enti
le.B
Mis
book
-to-
mar
ket
rati
o.B
Mis
win
sori
zed
atth
e1s
tpe
rcen
tile
and
99th
perc
enti
le.P
RC
issh
are
pric
e.T
UR
Nis
the
aver
age
mon
thly
turn
over
over
the
prev
ious
quar
ter.
VO
Lis
the
mon
thly
vola
tilit
yov
erth
epa
sttw
oye
ars.
SP50
0is
adu
mm
yva
riab
lefo
rS&
P50
0in
dex
mem
bers
hip.
RE
Tt
3,t
isth
ela
gged
thre
e-m
onth
retu
rn.
RE
Tt
12,t
3
isth
ela
gged
nine
-mon
thre
turn
prec
edin
gth
ebe
ginn
ing
ofth
equ
arte
r.P
anel
Apr
esen
tsth
eti
me-
seri
esm
ean,
med
ian,
max
imum
,min
imum
,and
stan
dard
devi
atio
nof
the
quar
terl
ycr
oss-
sect
iona
lav
erag
es.P
anel
Bpr
esen
tsth
eti
me-
seri
esav
erag
eof
the
cros
s-se
ctio
nalc
orre
lati
ons
betw
een
inst
itut
iona
low
ners
hip
and
stoc
kch
arac
teri
stic
s.P
anel
Cpr
esen
tsth
eti
me-
seri
esav
erag
eof
the
cros
s-se
ctio
nalc
orre
lati
ons
betw
een
stoc
kch
arac
teri
stic
s.
901
-
Both SIO and LIO are positively correlated with size, firm age,
andS&P 500 dummy, and are negatively correlated with
volatility. However,LIO has stronger correlations with these
variables than SIO. Further,SIO is significantly positively
correlated with turnover, whereas LIO isuncorrelated with turnover.
This result suggests that short-term institutionscare more about
liquidity than long-term institutions. SIO has weakpositive
correlations with past returns, while the correlations betweenLIO
and past returns are virtually zero. In addition, LIO is
positivelycorrelated with book-to-market ratio and dividend yield,
while SIO is notcorrelated with these variables. Overall, the above
results suggest that thereexist systematic differences between
long- and short-term institutionalpreferences. We note that these
bivariate correlations should be interpretedwith caution because of
the strong correlations between firm characteristics(reported in
Panel C of Table 1). In the next section, we use a
multivariateregression analysis to study the preferences of both
short- and long-terminstitutions.
2. Preferences of Short- and Long-Term Institutional
Investors
Prior literature [e.g., Falkenstein (1996), Del Guercio (1996),
Gompersand Metrick (2001), and Bennett, Sias, and Starks (2003)]
has examinedthe relation between institutional holdings and firm
characteristics. Theydocument that institutional investors prefer
certain firm characteristicssuch as size, share price, and
turnover. In this section, we explore whethershort- and long-term
institutional investors exhibit different preferencesfor firm
characteristics.
Following Gompers and Metrick (2001), we include three sets of
firmcharacteristics in our analysis. Firm size, age, dividend
yield, S&P 500index membership, and stock volatility are used
to proxy for prudence[e.g., Del Guercio (1996)]. Firm size, share
price, and stock turnover arerelated to liquidity and transactions
costs. Past returns, book-to-marketratios, and firm size have been
shown to predict future returns [e.g., Famaand French (1992) and
Jegadeesh and Titman (1993)]. For each quarterfrom the third
quarter of 1980 to the fourth quarter of 2003, we runthe following
cross-sectional regression of institutional ownership on theabove
firm characteristics:
INSTOWNi,t = 0 + 1MKTCAPi,t + 2AGEi,t + 3DPi,t + 4BMi,t+ 5PRCi,t
+ 6TURNi,t + 7VOLi,t + 8SP500i,t+ 9RETi,t3,t + 10RETi,t12,t3 +
ei,t
(5)
where INSTOWN is either total institutional ownership,
short-terminstitutional ownership, or long-term institutional
ownership.
902
The Review of Financial Studies / v 22 n 2 2009
-
Institutional Investors and Equity Returns
Panel A of Table 2 reports the time-series average of the
coefficientestimates. Since institutional ownership is extremely
persistent, we do notreport any statistical significance based on
the time-series of coefficientestimates (e.g., FamaMacBeth standard
errors). We follow Gompersand Metrick (2001) in reporting the
number of significant positive and
Table 2Determinants of institutional ownership and trading
Panel A: determinants of institutional ownership
Total Short-term Long-terminstitutional ownership institutional
ownership institutional ownership
Average [+significant, Average [+significant, Average
[+significant,coefficient significant] coefficient significant]
coefficient significant]
Market capitalization 0.045 [94, 0] 0.017 [93, 1] 0.009 [89,
0]Age 0.009 [61, 9] 0.006 [1, 63] 0.014 [81, 3]Dividend yield 0.316
[0, 92] 0.208 [0, 92] 0.005 [27, 30]Book-to-market 0.087 [94, 0]
0.023 [94, 0] 0.024 [93, 0]Price 0.072 [94, 0] 0.022 [94, 0] 0.021
[92, 0]Turnover 0.579 [93, 0] 0.419 [94, 0] 0.003 [34,
16]Volatility 0.020 [31, 14] 0.041 [56, 10] 0.017 [32, 8]S&P
500 0.035 [58, 16] 0.014 [30, 44] 0.040 [90, 0]RETt3,t 0.055 [2,
84] 0.014 [5, 61] 0.014 [0, 70]RETt12,t3 0.030 [2, 78] 0.001 [25,
21] 0.012 [1, 86]
Panel B: determinants of institutional trading
Total Short-term Long-terminstitutional trading institutional
trading institutional trading
Average [+significant, Average [+significant, Average
[+significant,coefficient significant] coefficient significant]
coefficient significant]
Market capitalization 100 0.067 [30, 12] 0.001 [17, 19] 0.041
[32, 13]Age 100 0.066 [7, 26] 0.008 [15, 11] 0.054 [21, 29]Dividend
yield 100 0.252 [23, 25] 0.358 [19, 18] 0.630 [26,
28]Book-to-market 100 0.010 [22, 17] 0.017 [17, 17] 0.055 [34,
19]Price 100 0.054 [17, 13] 0.027 [7, 18] 0.066 [31, 17]Turnover
0.018 [11, 45] 0.011 [11, 47] 0.002 [16, 16]Volatility 0.011 [17,
3] 0.001 [17, 7] 0.002 [12, 6]S&P 500 0.003 [12, 31] 0.000 [11,
16] 0.001 [24, 28]RETt 3,t 0.016 [83, 0] 0.013 [87, 0] 0.001 [12,
3]RETt 12,t 3 0.004 [40, 1] 0.001 [25, 5] 0.000 [12, 3]
This table summarizes the results of cross-sectional regressions
of institutional ownership andinstitutional trading on stock
characteristics. The sample period is from 1980:Q3 to
2003:Q4.Institutional holdings are obtained from Thomson Financial.
Stock characteristics are from the CRSPand COMPUSTAT database. An
institutional investor is classified as a short-term investor if
its past4-quarter turnover rate ranks in the top tertile. An
institutional investor is classified as a long-terminvestor if its
past 4-quarter turnover rate ranks in the bottom tertile. Age is
firm age measured asnumber of months since first return appears in
the CRSP database. Dividend yield is winsorized at the99th
percentile. Book-to-market ratio is winsorized at the 1st
percentile and 99th percentile. Turnoveris the average monthly
turnover over the previous quarter. Volatility is the monthly
volatility overthe past two years. S&P 500 is a dummy variable
for S&P 500 index membership. RETt 3,t is thelagged three-month
return. RETt 12,t 3 is the lagged nine-month return preceding the
beginning of thequarter. All variables except institutional
ownership, S&P 500 index membership, and lagged returnsare
expressed in natural logarithms. We estimate a cross-sectional
regression each quarter. We reportthe average regression
coefficient. In brackets, we report the number of coefficients that
are positive(and negative) and statistically significant at the 5%
level. All returns are in percent.
903
-
significant negative coefficients for 94 OLS regressions of
Equation (5).The first two columns report the results for total
institutional ownership,while the last four columns report the
results on short- and long-terminstitutional ownership.
Institutional investors as a whole show strong preference for
largerstocks and stocks with higher book-to-market value, higher
price, higherturnover, and lower dividend yield. For example, in
each of the 94 quarterlycross-sectional regressions, the
coefficient on market capitalization ispositive and statistically
significant at the 5% level. Further, institutionalinvestors show
some preference for older stocks, more volatile stocks,and stocks
that are members of the S&P 500 index. These results arebroadly
consistent with prior studies on institutional preference.
Consistentwith Gompers and Metrick (2001), we find that the
coefficients on pastreturns are significantly negative. Gompers and
Metrick conclude from thisresult that institutional investors are
not momentum investors. Bennett,Sias, and Starks (2003) and Sias
(2005), on the other hand, show thateven though institutional
holdings are negatively related to past returns,institutional
trading is positively related to past returns. Thus, they arguethat
institutional investors are momentum investors. We will examine
thisissue later when we present our results on institutional
trading.
We find both similarities and significant differences between
short- andlong-term institutions preferences for stock
characteristics. Both short-and long-term institutions prefer
larger stocks, and stocks with higherbook-to-market ratios and
share prices. However, short-term institutionsprefer younger firms,
while long-term institutions prefer older firms.Specifically, 63
out of 94 coefficients on age are significantly negativein
regressions of short-term institutional ownership, while 81 out of
94coefficients on age are significantly positive in regressions of
long-terminstitutional ownership. Further, long-term institutions
prefer S&P 500firms, while short-term institutions are
indifferent. In addition, short-term institutions show strong
preference for firms with lower dividendyield, while long-term
institutional ownership is not consistently related todividend
yield. These results suggest that short-term institutions are
lessconcerned about prudence than long-term institutions.
Although both short- and long-term institutions prefer stocks
withhigher turnover, short-term institutions have a much stronger
preferencefor turnover. This suggests that short-term institutions
care more aboutliquidity, presumably because they trade more
actively. Finally, bothshort- and long-term investors holdings are
negatively related to pastthree-month returns. However, their
relations to past one-year returns aresubstantially different.
Long-term institutional holdings are significantlynegatively
related to past one-year returns, while short-term
institutionalownership is not significantly related to past
one-year returns.
904
The Review of Financial Studies / v 22 n 2 2009
-
Institutional Investors and Equity Returns
To explore whether institutional investors are momentum
investorsalong the lines of Bennett, Sias, and Starks (2003), we
re-estimatethe regression Equation (5) by replacing institutional
ownership withchanges in institutional ownership as the dependent
variable. Panel B ofTable 2 reports the regression results. We
focus on the relation betweeninstitutional trading and past
returns. Results in Panel B indicate thatinstitutional investors as
a whole are momentum investors: institutionaltrading is
significantly positively related to past three-month or
one-yearreturns. Further analysis show that this result is
primarily driven byshort-term institutions. We find strong evidence
that short-term investorsare momentum traders. For example, the
regression coefficient on pastone-quarter returns is significantly
positive for 87 quarters, while none issignificantly negative. In
contrast, we find little evidence that long-terminstitutional
investors are momentum traders. The average point estimateson past
returns are virtually zero, and for most quarters they are
notstatistically significant.
3. The Impact of Institutional Ownership and Trading On Stock
Returns
3.1 The impact of institutional ownership on stock returns:
short-term versuslong-term institutionsGompers and Metrick (2001)
document a positive relation betweeninstitutional ownership and
next quarters stock returns. In this section weexamine the extent
to which the predictive ability of institutional ownershipis
attributable to short- and long-term institutions. Specifically,
for eachquarter, we run the following cross-sectional regression of
one-quarter-ahead (or one-year-ahead) stock returns on
institutional ownership andvarious firm characteristics:
RETi,t,t+3(RETi,t,t+12) = 0 + 1IOi,t + 2SIOi,t (or LIOi,t )+
3MKTCAPi,t + 4AGEi,t + 5DPi,t + 6BMi,t+ 7PRCi,t + 8TURNi,t +
9VOLi,t + 10SP500i,t+ 11RETi,t3,t + 12RETi,t12,t3 + ei,t
(6)
where RET i,t,t+3, and RET i,t,t+12 are one-quarter-ahead and
one-year-ahead stock returns, respectively. To make sure the
predictive ability ofinstitutional ownership is not driven by its
relation with other firm charac-teristics, we control for the same
set of firm characteristics that we use inour analysis of
institutional preference. Following Gompers and Metrick,we estimate
regression Equation (6) using weighted-least-squares, witheach firm
weighted by its log market capitalization. We estimate Equation(6)
quarter by quarter and use the Fama and MacBeth (1973) method
tocalculate standard errors for the time-series average of
coefficients.
Table 3 reports the time-series average of coefficient estimates
andthe associated p-values. The dependent variable is the
one-quarter-ahead return for the first set of three regressions,
while the dependent
905
-
Table 3Institutional ownership and future stock returns
Dependent variableRETt,t+3 Dependent variableRETt,t+12
Intercept 0.114 (0.01) 0.114 (0.01) 0.114 (0.01) 0.477 (0.01)
0.476 (0.01) 0.477 (0.01)IO 0.019 (0.01) 0.004 (0.61) 0.023 (0.01)
0.058 (0.02) 0.013 (0.60) 0.071 (0.01)SIO 0.054 (0.01) 0.142
(0.01)LIO 0.021 (0.05) 0.055 (0.07)BM 0.007 (0.01) 0.007 (0.01)
0.007 (0.01) 0.034 (0.01) 0.035 (0.01) 0.035 (0.01)MKTCAP 0.006
(0.01) 0.006 (0.01) 0.006 (0.01) 0.023 (0.01) 0.023 (0.01) 0.023
(0.01)VOL 0.023 (0.70) 0.025 (0.68) 0.024 (0.69) 0.161 (0.50) 0.166
(0.49) 0.162 (0.50)TURN 0.104 (0.01) 0.117 (0.01) 0.108 (0.01)
0.396 (0.01) 0.442 (0.01) 0.409 (0.01)PRC 0.008 (0.04) 0.008 (0.04)
0.008 (0.04) 0.020 (0.24) 0.020 (0.24) 0.020 (0.24)SP500 0.016
(0.01) 0.017 (0.01) 0.017 (0.01) 0.058 (0.01) 0.061 (0.01) 0.059
(0.01)RETt 3,t 0.004 (0.69) 0.004 (0.70) 0.004 (0.69) 0.109 (0.01)
0.109 (0.01) 0.109 (0.01)RETt 12,t 3 0.022 (0.01) 0.022 (0.01)
0.022 (0.01) 0.034 (0.01) 0.033 (0.01) 0.034 (0.01)AGE 0.002 (0.07)
0.003 (0.03) 0.002 (0.06) 0.003 (0.47) 0.004 (0.31) 0.003 (0.41)DP
0.026 (0.35) 0.022 (0.42) 0.024 (0.37) 0.128 (0.25) 0.116 (0.30)
0.126 (0.26)Avg. R2 0.080 0.081 0.080 0.078 0.079 0.078
This table summarizes the results of cross-sectional regressions
of one-quarter-ahead or one-year-aheadreturns on institutional
ownership and other stock characteristics. The sample period is
from 1980:Q3to 2003:Q4. Institutional holdings are obtained from
Thomson Financial. Stock characteristics are fromthe CRSP and
COMPUSTAT database. RETt,t+3 is one-quarter-ahead stock return.
RETt,t+12 is one-year-ahead stock return. IO is total institutional
ownership. SIO is short-term institutional ownership.LIO is
long-term institutional ownership. An institutional investor is
classified as a short-term investorif its past 4-quarter turnover
rate ranks in the top tertile. An institutional investor is
classified as along-term investor if its past 4-quarter turnover
rate ranks in the bottom tertile. MKTCAP is marketcapitalization.
Age is firm age measured as number of months since first return
appears in the CRSPdatabase. DP is dividend yield. DP is winsorized
at the 99th percentile. BM is book-to-market ratio.BM is winsorized
at the 1st percentile and 99th percentile. PRC is share price. TURN
is the averagemonthly turnover over the previous quarter. VOL is
the monthly volatility over the past two years.SP500 is dummy
variable for S&P 500 index membership. RETt 3,t is the lagged
three-month return.RETt 12,t 3 is the lagged nine-month return
preceding the beginning of the quarter. All variablesexcept
institutional ownership, S&P 500 index membership, and stock
returns are expressed in naturallogarithms. We use the Fama and
MacBeth (1973) methodology and report the time-series
averageregression coefficients. Numbers in parentheses are p-values
based on NeweyWest standard errors.Regression coefficients on
institutional ownership that are statistically significant at the
5% levels arein bold. All returns are in percent.
variable is the one-year-ahead return for the second set of
regressions. Inthe regression of one-year-ahead returns, the
residuals will be seriallycorrelated because the dependent variable
is overlapped. We reportp-values on the basis of the NeweyWest
(1987) standard errors toaccount for this autocorrelation. Within
each set of regressions, we firstinclude only total institutional
ownership, and then add short-term orlong-term institutional
ownership to the regressions.
We find strong evidence that institutional ownership forecasts
one-quarter-ahead returns. The average coefficient on IO is 0.019
and isstatistically significant at the 1% level. When we include
both total IO andSIO in the regression, the predictive power of IO
is completely subsumedby SIO. The average coefficient for SIO is
0.054 and statistically significantat the 1% level. This result is
also economically significant. A two-standarddeviation increase in
SIO is associated with an increase in next quartersstock return by
about 1.1%. After controlling for SIO, the point estimatefor IO
drops from 0.019 to 0.004, with a p-value of 0.61. By contrast,
when
906
The Review of Financial Studies / v 22 n 2 2009
-
Institutional Investors and Equity Returns
we include both IO and LIO in the regression, the average
coefficient onIO remains positive and significant, while the
average coefficient on LIOis actually negative and significant at
the 5% level.7
The results for one-year-ahead returns are similar. When used
alone,current quarter IO has significant predictive power for the
next yearsstock returns. However, after controlling for SIO, the
predictive abilityof IO disappears, while the average coefficient
on SIO is positive andhighly significant. Like the results on
one-quarter-ahead returns, themarginal effect of long-term
institutional ownership on one-year-aheadreturns is significantly
negative after controlling for total institutionalownership.
Overall, the results in this section indicate that institutional
ownershiphas strong predictive ability for both next quarters and
next yearsreturns. This result is consistent with Gompers and
Metrick (2001).More importantly, we show that this predictive
ability is almost entirelydriven by short-term institutions. By
contrast, long-term institutionalownership does not have
incremental predictive power for future stockreturns.
3.2 Demand shock versus informational advantageGompers and
Metrick (2001) argue that two forces may be drivingthe positive
relation between institutional ownership and future
returns:institutions either provide persistent demand shocks or
they have aninformational advantage. To disentangle these two
effects, they decomposethe current quarter institutional ownership
(IOt ) into lagged institutionalownership (IOt1) and the change in
institutional ownership (IOt ).If the predictive ability of
institutional ownership is due to demandshock, given that
institutional holdings are quite stable, one wouldexpect that IOt1
has a stronger predictive power. If institutionalinvestors have an
informational advantage, then IOt should be a betterpredictor.
To examine the sources of predictive ability of short-term
institutionalownership, we decompose the current institutional
holdings into laggedholdings and changes in holdings for both
short- and long-term investors:SIOt1,SIOt , LIOt1, and LIOt . Next,
for each quarter we run the
7 One should not interpret this result as evidence that
long-term institutional ownership has negativepredictive ability
for future returns. We note that the coefficient on LIO captures
only the marginal effectof long-term institutional ownership. In
particular, the total institutional ownership, which also
containslong-term institutional ownership, has a positive relation
with future stock returns, Therefore, the totaleffect of long-term
institutional ownership is likely indistinguishable from zero.
907
-
following cross-sectional regression:
RETi,t,t+3(RETi,t,t+12) = 0 + 1SIOi,t1 + 2LIOi,t1 +
3SIOi,t+4LIOi,t + 5MKTCAPi,t + 6AGEi,t + 7DPi,t+8BMi,t + 9PRCi,t +
10TURNi,t + 11VOLi,t
+12SP500i,t + 13RETi,t3,t + 14RETi,t12,t3 + ei,t(7)
Table 4 reports the time-series average of coefficients and
associatedp-values calculated on the basis of the NeweyWest
standard errors.When we include only ownership variables in the
regression, SIOt stronglypredicts future returns, while the average
coefficient on LIOt is statistically
Table 4Short-term institutional investors, long-term
institutional investors, and future stock returns
Dependent variableRETt,t+3 Dependent variableRETt,t+12
Intercept 0.115 (0.01) 0.119 (0.01) 0.479 (0.01) 0.472
(0.01)SIOt 0.058 (0.01) 0.157 (0.01)LIOt 0.004 (0.68) 0.024
(0.45)SIOt 1 0.043 (0.01) 0.135 (0.01)LIOt 1 0.016 (0.10) 0.032
(0.32)SIOt 0.062 (0.01) 0.192 (0.01)LIOt 0.021 (0.12) 0.006
(0.87)BM 0.007 (0.01) 0.007 (0.01) 0.035 (0.01) 0.036 (0.01)MKTCAP
0.006 (0.01) 0.006 (0.01) 0.023 (0.01) 0.023 (0.01)VOL 0.026 (0.67)
0.036 (0.56) 0.168 (0.48) 0.150 (0.53)TURN 0.117 (0.01) 0.114
(0.01) 0.441 (0.01) 0.436 (0.01)PRC 0.008 (0.04) 0.008 (0.03) 0.020
(0.24) 0.020 (0.25)SP500 0.017 (0.01) 0.017 (0.01) 0.060 (0.01)
0.059 (0.01)RETt 3,t 0.004 (0.70) 0.002 (0.84) 0.110 (0.01) 0.108
(0.01)RETt 12,t 3 0.022 (0.01) 0.021 (0.01) 0.033 (0.01) 0.033
(0.01)AGE 0.002 (0.04) 0.002 (0.05) 0.004 (0.33) 0.004 (0.40)DP
0.022 (0.42) 0.022 (0.42) 0.119 (0.29) 0.120 (0.30)Avg. R2 0.081
0.083 0.078 0.080
This table summarizes the results of cross-sectional regressions
of one-quarter-ahead orone-year-ahead returns on short-term and
long-term institutional ownership, and otherstock characteristics.
The sample period is from 1980:Q3 to 2003:Q4. Institutional
holdingsare obtained from Thomson Financial. Stock characteristics
are from the CRSP andCOMPUSTAT database. RETt,t+3 is
one-quarter-ahead stock return. RETt,t+12 is one-year-ahead stock
return. IO is total institutional ownership. SIO is short-term
institutionalownership. LIO is long-term institutional ownership.
An institutional investor is classified asa short-term investor if
its past 4-quarter turnover rate ranks in the top tertile. An
institutionalinvestor is classified as a long-term investor if its
past 4-quarter turnover rate ranks in thebottom tertile. MKTCAP is
market capitalization. Age is firm age measured as numberof months
since first return appears in the CRSP database. DP is dividend
yield. DP iswinsorized at the 99th percentile. BM is book-to-market
ratio. BM is winsorized at the 1stpercentile and 99th percentile.
PRC is share price. TURN is the average monthly turnoverover the
previous quarter. VOL is the monthly volatility over the past two
years. SP500 isdummy variable for S&P 500 index membership.
RETt3,t is the lagged three-month return.RETt12,t3 is the lagged
nine-month return preceding the beginning of the quarter.
Allvariables except institutional ownership, S&P 500 index
membership, and stock returns areexpressed in natural logarithms.
We estimate cross-sectional regressions for each quarter. Weuse the
Fama and MacBeth (1973) methodology and report the time-series
average regressioncoefficients. Numbers in parentheses are p-values
based on NeweyWest standard errors.Regression coefficients on
institutional ownership and trading that are statistically
significantat the 5% levels are in bold. All returns are in
percent.
908
The Review of Financial Studies / v 22 n 2 2009
-
Institutional Investors and Equity Returns
insignificant. This result is consistent with those reported in
Table 4 that thepredictive ability of total institutional ownership
is driven by short-terminstitutions.
When we include both lagged institutional holdings and changes
ininstitutional holdings, the average coefficient on SIOt1 is still
statisticallysignificant. This result suggests that demand shocks
impact stockreturns. More importantly, the average coefficient on
SIOt is positiveand statistically significant at the 1% level,
suggesting that short-terminstitutions are informed. Since our
results are obtained after controllingfor various stock
characteristics including size, book-to-market, and pastreturns,
they cannot be explained by short-term institutions
followingcertain investment styles that have been shown to explain
cross-sectionalstock returns. In particular, our results are not
driven by the momentumeffect. In contrast, neither LIOt1 nor LIOt
is significantly relatedto future stock returns. The results for
one-year-ahead returns arequalitatively similar. Short-term
institutions holdings and trading bothforecast one-year-ahead
returns, while neither holdings nor trading bylong-term
institutions has any predictive power for next years returns.
Our results on short-term institutions are economically
significant.A two-standard-deviation change in short-term
institutional trading isassociated with a change in next quarters
return by 0.53%, and isassociated with a change in next years
return by 1.63%. Overall, theseresults suggest that short-term
institutions are informed. Moreover, theyare better informed than
long-term institutions.
3.3 The information content of short-term institutional trading:
small/big,value/growthIf short-term institutions possess superior
information about futurereturns, their informational advantage
should be greater for smaller firmsand firms with more growth
opportunities. In general, these firms facemore uncertainty and
their values are more difficult to evaluate [e.g.,Wermers (1999)
and Sias (2004)]. To test this implication, we divide allsample
stocks into small/large and value/growth categories.
Specifically,each quarter, a firm is classified as a small firm if
its market capitalizationis lower than the NYSE median. Otherwise,
it is considered a big firm.Similarly, a firm is classified as a
growth firm if its book-to-market ratio isless than the
cross-sectional median; otherwise, it is a value firm.
We then re-estimate regression Equation (7) for small/large
andvalue/growth firms separately. If the predictive ability of
short-terminstitutional trading reflects superior information, we
would expect SIOtto have a stronger predictive power for small and
growth stocks than forlarge and value stocks.
Panel A of Table 5 reports the regression results for
small/large stocks.Again we examine both one-quarter-ahead and
one-year-ahead returns.
909
-
We focus our discussion on the coefficients on institutional
trading becauseit captures the informational advantage better. For
one-quarter-aheadreturns, the average coefficient on SIOt for small
stocks is 0.087, abouttwice as large as that for large stocks
(0.047). Since the standard deviationsof SIOt for small and large
stocks are very similar (the average cross-sectional standard
deviation for SIOt is 4.1% for small stocks and 3.9%for large
stocks), the above estimates for regression coefficients imply
thatthe marginal effect of SIOt is about twice as big for small
stocks asthat for large stocks. In addition, although the
coefficient estimates onSIOt are statistically significant for both
small and large stocks, thep-value is smaller for small stocks
(p-value = 0.01) than for large stocks(p-value = 0.03). Similar
results hold for one-year-ahead returns. Bothsmall and large firms
have statistically significant coefficients on SIOt ,but the point
estimate is again bigger for small stocks than for large
stocks(0.27 versus 0.19). Overall, we find evidence that the
predictive power ofshort-term institutional trading is stronger for
small stocks than for largestocks.
Consistent with our earlier results, we find no evidence that
long-terminstitutional trading predicts future returns for either
small or large stocks.For one-quarter-ahead returns, the average
coefficients for LIOt 1 andLIOt are insignificant for both small
and large firms. The point estimatesfor LIOt are actually negative.
For one-year-ahead returns, althoughthe coefficient on LIOt 1 is
positive and statistically significant for smallstocks, the average
coefficient on LIOt is statistically insignificant forboth small
and large stocks.
Table 5Short-term institutional investors, long-term
institutional investors, and future stock returns:small/large and
value/growth
Panel A: small/large stocks
Dependent variableRETt,t+3 Dependent variableRETt,t+12Small
Large Small Large
Intercept 0.142 (0.01) 0.052 (0.05) 0.589 (0.01) 0.180
(0.02)SIOt 1 0.057 (0.01) 0.031 (0.03) 0.200 (0.01) 0.046
(0.42)LIOt 1 0.024 (0.06) 0.017 (0.09) 0.113 (0.01) 0.039
(0.24)SIOt 0.087 (0.01) 0.047 (0.03) 0.266 (0.01) 0.191 (0.01)LIOt
0.029 (0.07) 0.008 (0.64) 0.030 (0.54) 0.009 (0.86)BM 0.011 (0.01)
0.011 (0.06) 0.051 (0.01) 0.055 (0.01)MKTCAP 0.009 (0.01) 0.001
(0.35) 0.34 (0.01) 0.005 (0.55)VOL 0.011 (0.85) 0.119 (0.19) 0.031
(0.88) 0.211 (0.52)TURN 0.115 (0.01) 0.057 (0.06) 0.510 (0.01)
0.120 (0.35)PRC 0.01 (0.07) 0.002 (0.47) 0.021 (0.36) 0.002
(0.85)SP500 0.011 (0.06) 0.003 (0.28) 0.047 (0.02) 0.013 (0.17)RETt
3,t 0.005 (0.55) 0.002 (0.88) 0.119 (0.01) 0.120 (0.01)RETt 12,t 3
0.022 (0.01) 0.032 (0.01) 0.033 (0.02) 0.061 (0.01)AGE 0.002 (0.13)
0.001 (0.79) 0.000 (0.98) 0.001 (0.95)DP 0.013 (0.11) 0.005 (0.82)
0.072 (0.06) 0.046 (0.60)Avg. R2 0.075 0.141 0.071 0.135
910
The Review of Financial Studies / v 22 n 2 2009
-
Institutional Investors and Equity Returns
Table 5(Continued)
Panel B: value/growth stocks
Dependent variableRETt,t+3 Dependent variableRETt,t+12Value
Growth Value Growth
Intercept 0.084 (0.01) 0.118 (0.01) 0.436 (0.01) 0.416
(0.01)SIOt 1 0.016 (0.27) 0.077 (0.01) 0.045 (0.40) 0.223
(0.01)LIOt1 0.013 (0.19) 0.044 (0.01) 0.014 (0.67) 0.180 (0.01)SIOt
0.040 (0.03) 0.103 (0.01) 0.119 (0.07) 0.352 (0.01)LIOt 0.029
(0.07) 0.029 (0.28) 0.041 (0.44) 0.087 (0.11)BM 0.015 (0.01) 0.002
(0.84) 0.034 (0.09) 0.020 (0.63)MKTCAP 0.004 (0.01) 0.007 (0.01)
0.019 (0.01) 0.023 (0.01)VOL 0.036 (0.56) 0.047 (0.46) 0.162 (0.49)
0.122 (0.59)TURN 0.104 (0.01) 0.116 (0.01) 0.406 (0.01) 0.445
(0.01)PRC 0.010 (0.01) 0.006 (0.15) 0.027 (0.14) 0.007 (0.65)SP500
0.015 (0.01) 0.015 (0.01) 0.057 (0.01) 0.045 (0.01)RETt 3,t 0.009
(0.37) 0.014 (0.16) 0.117 (0.01) 0.120 (0.01)RETt 12,t3 0.029
(0.01) 0.022 (0.01) 0.043 (0.02) 0.041 (0.01)AGE 0.003 (0.03) 0.002
(0.33) 0.002 (0.57) 0.002 (0.75)DP 0.012 (0.12) 0.057 (0.15) 0.034
(0.28) 0.245 (0.12)Avg. R2 0.092 0.077 0.094 0.067
This table summarizes the results of cross-sectional regressions
of one-quarter-ahead orone-year-ahead returns on short-term and
long-term institutional ownership, and otherstock characteristics
for small/large, and value/growth stocks separately. The
sampleperiod is from 1980:Q3 to 2003:Q4. Institutional holdings are
obtained from ThomsonFinancial. Stock characteristics are from the
CRSP and COMPUSTAT database. Largestocks have market capitalization
greater than that of the median NYSE stock. Smallstocks have market
capitalization less than that of the median NYSE stock. RETt,t+3is
one-quarter-ahead stock return. RETt,t+12 is one-year-ahead stock
return. IO is totalinstitutional ownership. SIO is short-term
institutional ownership. LIO is long-terminstitutional ownership.
An institutional investor is classified as a short-term investor if
itspast 4-quarter turnover rate ranks in the top tertile. An
institutional investor is classifiedas a long-term investor if its
past 4-quarter turnover rate ranks in the bottom tertile.MKTCAP is
market capitalization. Age is firm age measured as number of months
sincefirst return appears in the CRSP database. DP is dividend
yield. DP is winsorized atthe 99th percentile. BM is book-to-market
ratio. BM is winsorized at the 1st percentileand 99th percentile.
PRC is share price. TURN is the average monthly turnover overthe
previous quarter. VOL is the monthly volatility over the past two
years. SP500 isdummy variable for S&P 500 index membership.
RETt3,t is the lagged three-monthreturn. RETt 12,t 3 is the lagged
nine-month return preceding the beginning of thequarter. All
variables except institutional ownership, S&P 500 index
membership, andstock returns are expressed in natural logarithms.
We estimate cross-sectional regressionsfor each quarter. We use the
Fama and MacBeth (1973) methodology and report thetime-series
average regression coefficients. Numbers in parentheses are
p-values basedon NeweyWest standard errors. Regression coefficients
on institutional ownership andtrading that are statistically
significant at the 5% levels are in bold. All returns are
inpercent.
Panel B of Table 5 report the results for value/growth stocks.
Sincethe standard deviations of SIOt are nearly identical between
value andgrowth stocks (4% for value stocks and 4.1% for growth
stocks), we candirectly compare the magnitude of the coefficients
on SIOt across valueand growth stocks. Results in Panel B indicate
that for one quarter-ahead-returns, the coefficient on SIOt is more
than twice as large for growthstocks as that for value stocks
(0.103 versus 0.04). Further, the coefficient
911
-
on SIOt is statistically significant at the 1% level for growth
stocks, andonly at the 5% level for value stocks.
The results for one-year-ahead returns reveal even greater
differencesbetween growth and value stocks. The coefficient on
SIOt1 is statisticallysignificant for growth stocks but
insignificant for value stocks. Moreimportantly, the coefficient on
SIOt is statistically significant at the 1%level for growth stocks,
but is insignificant at the 5% level (p-value = 0.07)for value
stocks. In addition, the coefficient on SIOt is 0.352 for
growthstocks, about three times as high as that for value stocks
(0.119).
In summary, results in Table 5 indicate that short-term
institutionaltrading has stronger predictive power for small and
growth stocks thanfor large and value stocks. These results are
consistent with the hypothesisthat short-term institutional trading
predicts future returns because theyhave an informational
advantage.
3.4 Portfolio approachThus far we have focused on a
cross-sectional regression approach toexamine the effect of
institutional ownership and trading on futurereturns. This approach
allows us to compare with prior studies [e.g.,Gompers and Metrick
(2001)], and to readily control for known predictorsof future
returns such as size, book-to-market, and past returns. Togauge the
robustness of our results, in this section we use a
portfolioapproach.
Specifically, we construct investment portfolios based on short-
andlong-term institutional trading. At the end of each quarter, we
rank allsample stocks on the basis of their current-quarter changes
in short-/long-term institutional ownership, and sort them into
five portfolios. We holdthese portfolios for one year and report
the cumulative value-weightedholding period returns on the
portfolio of stocks with the largest increasein short-/long-term
institutional holding (Q5) and returns on the portfolioof stocks
with the largest decrease in short-/long-term institutional
holding(Q1). We also report the return on a zero-investment
strategy that is longin portfolio Q5 and short in portfolio Q1,
where Q5 and Q1 are formedevery quarter as above, and held for one
year. In addition to raw returns,we also report the Daniel et al.
(1997) (DGTW) benchmark-adjustedreturns. DGTW benchmark-adjusted
returns allow us to control for thesize, book-to-market, and
momentum effect.
Table 6 reports the results on portfolios sorted by short- and
long-term institutional trading. The portfolio of stocks with the
largestincrease in short-term institutional holdings earns higher
returns than theportfolio of stocks with the largest decrease in
short-term institutionaltrading. For the raw return, the average
quarterly return on thezero-investment strategy Q5 Q1 is 0.53%
(t-statistics = 3.16) over thefour quarters after portfolio
formation. This average return difference
912
The Review of Financial Studies / v 22 n 2 2009
-
Institutional Investors and Equity Returns
Table 6Returns on portfolios sorted by changes in short- and
long-term institutional ownership
QuartersQuarterly t + 1 t + 1 t + 1Average t + 1 through t + 2
through t + 3 through t + 4
Short-term institutional trading portfoliosQ5 3.62 3.61 6.91
10.61 14.88Q1 3.09 2.97 5.86 8.74 12.72Q5 Q1 0.53 0.64 1.05 1.87
2.16
(3.16) (2.02) (2.42) (3.16) (2.77)Q5 Q1 (DGTW 0.41 0.42 0.62
1.24 1.62
adjusted) (3.34) (1.73) (1.86) (2.74) (2.82)Long-term
institutional trading portfolios
Q5 3.21 2.91 5.90 9.08 13.27Q1 3.40 3.08 6.55 10.18 14.02Q5 Q1
0.19 0.17 0.64 1.09 0.75
(1.20) (0.49) (1.52) (2.25) (1.41)Q5 Q1 (DGTW 0.03 0.04 0.35
0.55 0.16
adjusted) (0.29) (0.16) (1.12) (1.52) (0.41)
This table reports the returns on portfolios sorted by the
quarterly change in short- and long-terminstitutional ownership.
The sample period is from 1980:Q3 to 2003:Q4. Institutional
holdings areobtained from CDA/Spectrum. Stock returns are from the
CRSP. An institutional investor is classified asa short-term
investor if its past 4-quarter portfolio turnover rate ranks in the
top tertile. An institutionalinvestor is classified as a long-term
investor if its past 4-quarter turnover rate ranks in the
bottomtertile. Each quarter, we group all stocks available in
CDA/Spectrum into 5 portfolios based on theirrankings on change in
short- and long-term institutional ownership, respectively.
Portfolio Q5 containsstocks that experience the largest increase in
institutional ownership. Portfolio Q1 contains stocks
thatexperience the largest decrease in institutional ownership. For
each of portfolios Q5 and Q1, we reporttheir value-weighted
cumulative quarterly returns up to 4 quarters after the portfolio
formation. We alsoreport the average quarterly returns on an
investment strategy that is long in Q5 and short in Q1. Wereport
the time-series means of both raw returns as well as the Daniel et
al. (1997) (DGTW) benchmarkadjusted returns. The returns are in
percent. Numbers in parentheses are t-statistics. Return
differencesthat are statistically significant at 10% are in bold.
All returns are in percent.
decreases to 0.41% using DGTW-adjusted returns, but is still
statisticallysignificant (t-statistics = 3.34). Over the one year
after portfolio formation,the cumulative return difference between
Q5 and Q1 is 2.16% (t-statistics = 2.77) using raw returns, and is
1.62% using DGTW-adjustedreturns (t-statistics = 2.82).
Consistent with our earlier results, we find no evidence that
long-term institutional investors have stock-picking ability.
Whether measuredby raw returns or DGTW-adjusted returns, theres no
significantspread between Q5 and Q1 portfolios. Indeed, the return
differencesbetween Q5 and Q1 are actually negative for long-term
institutionalinvestors.
In summary, results in Table 6 show a significant difference
betweenlong-term and short-term investors. Short-term institutions
tradingstrongly predicts future returns while long-term
institutions trading doesnot. These results are consistent with
those reported in Table 4 and suggestthat short-term institutional
investors are better informed than long-terminstitutional
investors.
913
-
3.5 Institutional trading and future earnings newsWe have shown
that short-term institutional trading contains informationabout
future stock returns. To provide more direct evidence that
short-terminstitutions possess private information that is useful
in predicting futurereturns, we examine in this section the
relation between institutional tradingand future earnings news. We
examine both earnings announcementabnormal returns and earnings
surprises.
We obtain analysts consensus quarterly earnings forecast and
actualearnings from I/B/E/S. We obtain quarterly earnings
announcement datesfrom COMPUSTAT. The earnings announcement
abnormal return isdefined as the cumulative market-adjusted return
over a 3-day window[1,+1] around the earnings announcement date.
The earnings surprise isdefined as the difference between reported
earnings and consensus analystsearnings forecast divided by the
stock price of the previous quarter. Eachquarter, we group stocks
into five portfolios based on short-term/long-term institutional
trading. Specifically, portfolio Q5 (Q1) contains stocksfor which
the quarterly institutional ownership has increased (decreased)the
most. For each portfolio at each quarter, we then calculate the
medianearnings announcement abnormal returns and earnings surprises
over eachof the next four quarterly earnings announcements.
Panel A of Table 7 reports the time-series average of the median
earningsannouncement abnormal returns for portfolios Q5 and Q1, as
well as theearnings announcement return difference between these
two portfolios.Results in Table 7 indicate that short-term
institutional trading is positivelyrelated to future earnings
abnormal returns. Portfolio Q5 has an earningsannouncement abnormal
return 94 basis points higher than portfolio Q1in the first
quarter, and this difference is statistically significant at the1%
level (t-statistics = 17.38). The same pattern holds for the next
threequarters. Although the magnitude of the difference is smaller
for the nextthree quarters, it is still statistically
significant.
To the extent that institutions might trade on the basis of past
earningsnews, the above results may be influenced by the
well-documentedphenomenon of earnings momentum. To control for
earnings momentum,each quarter we divide all stocks into tertile
portfolios based on the currentquarters earnings announcement
abnormal return. Next, we calculate themedian abnormal return
difference between the buy and sell portfoliosaround subsequent
quarterly earnings announcements within each currentearnings
announcement abnormal return tertile. We then report the
time-series average of these return differences across the current
earningsannouncement abnormal return tertiles to stratify the
earnings momentumeffect. After adjusting for earnings momentum, we
find that the earningsannouncement return difference between Q5 and
Q1 remains significantand positive for the first three quarters.
The adjusted return differenceis about 53 basis points annualized,
comparable in magnitude to Baker
914
The Review of Financial Studies / v 22 n 2 2009
-
Institutional Investors and Equity Returns
Table 7Institutional trading and future earnings news
Panel A: cumulative market-adjusted return over [1,+1] around
earnings announcement (%)
Quarters
t + 1 t + 2 t + 3 t + 4Short-term institutional trading
portfolios
Q5 0.586 0.123 0.131 0.067Q1 0.351 0.006 0.018 0.000Q5 Q1 0.938
0.128 0.113 0.066
(17.38) (3.66) (3.23) (1.83)Q5 Q1 (earnings mom. adj.) 0.133
0.121 0.071 0.006
(4.15) (3.63) (2.04) (0.20)Long-term institutional trading
portfolios
Q5 0.074 0.075 0.076 0.055Q1 0.096 0.089 0.062 0.060Q5 Q1 0.022
0.015 0.014 0.005
(0.67) (0.45) (0.47) (0.15)Q5 Q1 (earnings mom. adj.) 0.014
0.008 0.003 0.005
(0.53) (0.30) (0.11) (0.15)
Panel B: earnings surprises (%)
t + 1 t + 2 t + 3 t + 4Short-term institutional trading
portfoliosQ5 0.047 0.003 0.003 0.004Q1 0.029 0.024 0.019 0.017Q5 Q1
0.076 0.027 0.016 0.013
(2.03) (7.02) (5.36) (5.42)Long-term institutional trading
portfoliosQ5 0.006 0.011 0.011 0.009Q1 0.014 0.013 0.009 0.008Q5 Q1
0.008 0.001 0.002 0.000
(1.31) (0.44) (0.80) (0.15)
This table reports the earnings announcement abnormal returns
and earnings surprises byinstitutional trading portfolios. The
sample period is from 1980:Q3 to 2003:Q4 for Panel A, andis 1984:Q1
to 2003:Q4 for Panel B. Institutional holdings are obtained from
Thomson Financial.Stock characteristics are from the CRSP and
COMPUSTAT database. An institutional investoris classified as a
short-term investor if its past 4-quarter turnover rate ranks in
the top tertile. Aninstitutional investor is classified as a
long-term investor if its past 4-quarter turnover rate ranksin the
bottom tertile. The earnings announcement abnormal return is
calculated for the three daysaround the earnings announcement date.
The earnings surprise is calculated as the differencebetween actual
earnings and consensus analyst forecast divided by the stock price.
Earnings dataare from I/B/E/S. Earnings announcement dates are
obtained from COMPUSTAT. We dividestocks into quintile portfolios
based on the changes of quarterly institutional ownership.
PortfolioQ5 contains stocks that experience the largest increase in
institutional ownership. Portfolio Q1contains stocks that
experience the largest decrease in institutional ownership. We
report thetime-series mean of cross-sectional median values for
these portfolios. Panel A reports the resultsfor earnings
announcement abnormal returns, and Panel B reports those for
earnings surprises.Numbers in parentheses are t-statistics.
Differences in returns or earnings surprises that arestatistically
significant at the 10% level are in bold.
et al. (2004) who examine the relation between mutual fund
trading andsubsequent earnings announcement returns.
In contrast to findings on short-term institutions, we find
noevidence that long-term institutional trading is related to
futureearnings announcement abnormal returns. The difference in
earningsannouncement abnormal returns between Q5 and Q1 is
indistinguishable
915
-
from zero across all four quarters whether we control for
earningsmomentum or not. This result is consistent with our earlier
finding thatlong-term institutional trading does not forecast
future stock returns.
Next, we examine if institutional trading is related to future
earningssurprises. Panel B reports the time-series average of
median earningssurprises for portfolios Q5 and Q1, as well as the
difference between Q5and Q1. Similar to the results for earnings
announcement returns, stocksfor which short-term institutional
ownership increases the most experiencesignificantly higher
earnings surprises (more positive or less negative) thanthose
stocks for which short-term institutional ownership decreases
themost. This difference is statistically significant for all four
quarters afterportfolio formation. By contrast, we find no evidence
that stocks thatlong-term institutions buy or sell exhibit
significantly different earningssurprises in any of the subsequent
four quarters.8
In summary, we show in this section that short-term
institutionaltrading is positively associated with both future
earnings surprises andabnormal returns around subsequent earnings
announcements. Theseresults provide more direct evidence that
short-term institutions possessprivate information about future
firm value. Consistent with our results onreturns, we find little
evidence that long-term institutional trading is relatedto either
future earnings surprises or earnings announcement
abnormalreturns.
3.6 Evidence on long-run returnsWe have shown that short-term
institutional trading forecasts futurereturns and contains
information about future earnings news, whichsuggests that
short-term institutions are informed. An alternativeexplanation for
our results is that short-term institutional investorspressure
corporate managers to maximize short-run profits at the expenseof
long-run firm value. Porter (1992) argues that the short-term
focusby institutional investors forces managers to be overly
concerned withmeasures of short-term performance such as quarterly
earnings. Bushee(2001) shows that transient institutions exhibit
strong preferences forcorporations with more value in expected
near-term earnings and lessin long-run value. Further, Bushee
(1998) finds that firms with highertransient institutional
ownership are more likely to underinvest in long-term, intangible
projects such as R&D to reverse an earnings decline.
While the short-term pressure hypothesis might explain the
short-runpredictive ability of short-term institutional ownership
and trading, it
8 Ke and Ramalingegowda (2005) also examine the relation between
changes in institutional ownership andfuture earnings news. They
focus on the revision of long-term earnings forecasts. They find no
evidencethat dedicated institutions have private information about
long-term earnings, and transient institutionsseem to possess
information about long-term earnings that will be reflected in
short-term stock prices.
916
The Review of Financial Studies / v 22 n 2 2009
-
Institutional Investors and Equity Returns
also predicts long-run price reversal for stocks held or traded
by short-term institutions. Our results for one-year-ahead returns
(reported inTables 4 and 5) are inconsistent with that hypothesis:
changes in short-term institutional holdings strongly predict next
years return. However,it is possible that one year is not long
enough for prices to revert to theirfundamental values. Therefore,
in Table 8 we examine the relation betweeninstitutional trading and
future stock returns up to three years. Specifically,we re-estimate
regression Equation (7) replacing the dependent variable bythe
two-year holding period return starting from one-year from the
current
Table 8Short-term institutional investors, long-term
institutional investors, and long-run stockreturns
Dependent variableRETt+12,t+36
Intercept 0.569 (0.01) 0.545 (0.01)SIOt 0.076 (0.18)LIOt 0.012
(0.82)SIOt1 0.055 (0.34)LIOt1 0.037 (0.42)SIOt 0.018 (0.74)LIOt
0.007 (0.84)BM 0.072 (0.01) 0.068 (0.01)MKTCAP 0.028 (0.01) 0.027
(0.01)VOL 0.405 (0.26) 0.392 (0.28)TURN 0.359 (0.08) 0.347
(0.10)PRC 0.030 (0.16) 0.032 (0.15)SP500 0.048 (0.12) 0.046
(0.02)RETt3,t 0.036 (0.29) 0.040 (0.27)RETt12,t3 0.023 (0.39) 0.022
(0.42)AGE 0.003 (0.74) 0.003 (0.81)DP 0.172 (0.45) 0.134 (0.56)Avg.
R2 0.050 0.050
This table summarizes the results of cross-sectional regressions
of long-run stock returnson short-term and long-term institutional
ownership, and other stock characteristics. Thesample period is
from 1980:Q3 to 2002:Q4. The institutional holdings are obtained
fromThomson Financial. The stock characteristics are from the CRSP
and COMPUSTATdatabase. RETt+12,t+36 is the 2-year cumulative return
beginning one year from thereport date. IO is total institutional
ownership. SIO is short-term institutional ownership.LIO is
long-term institutional ownership. An institutional investor is
classified as a short-term investor if its past 4-quarter turnover
rate ranks in the top tertile. An institutionalinvestor is
classified as a long-term investor if its past 4-quarter turnover
rate ranksin the bottom tertile. MKTCAP is market capitalization.
Age is firm age measured asnumber of months since first return
appears in the CRSP database. DP is dividend yield.DP is winsorized
at the 99th percentile. BM is book-to-market ratio. BM is
winsorizedat the 1st percentile and 99th percentile. PRC is share
price. TURN is the averagemonthly turnover over the previous
quarter. VOL is the monthly volatility over thepast two years.
SP500 is dummy variable for S&P 500 index membership. RETt 3,t
isthe lagged three-month return. RETt 12,t 3 is the lagged
nine-month return precedingthe beginning of the quarter. All
variables except institutional ownership, S&P 500index
membership, and stock returns are expressed in natural logarithms.
We estimatecross-sectional regressions for each quarter. We use the
Fama and MacBeth (1973)methodology and report the time-series
average regression coefficients. Numbers inparentheses are p-values
based on NeweyWest standard errors. Regression coefficientson
institutional ownership and trading that are statistically
significant at the 5% levelsare in bold. All returns are in
percent.
917
-
quarter. In this analysis, since the dependent variable is
overlapped, we useNeweyWest standard errors to account for serial
correlation of residuals.
Table 8 presents the regression results. If there is long-run
price reversal,we would expect the coefficients on SIOt and SIOt to
be significantlynegative. We find no evidence of long-run price
reversal. Indeed, thecoefficients on SIOt and SIOt are all positive
and insignificant. Thissuggests that the stocks held or bought by
short-term institutions donot under-perform in the long run. These
results confirm that short-terminstitutional trading predicts
future stock returns not because of the short-term pressure they
create for corporate managers, but because they havean
informational advantage. Thus, this articles main findings cannot
beexplained by the short-term pressure hypothesis.
Although long-term institutions have no ability to predict
short-termreturns, it is possible that they have superior long-term
information (as wediscussed in the introduction). Results in Table
8 are not consistent withthis hypothesis. Neither holdings nor
trading by long-term institutionsforecast long-run returns.
Specifically, the coefficients on LIOt 1 andLIOt are not
statistically significant at any conventional levels.
In summary, we show in this section that our results are not
explainedby the short-term pressure hypothesis. While we might have
overlookedother alternative explanations, the fact that we do not
find long-run pricereversal likely rules out other
non-information-based explanations as well:if price movements
result from something other than information, wewould expect to
observe subsequent return reversals.
Table 9Short-term institutional investors, long-term
institutional investors, and future stock returns:subperiods
Panel A: 19801991
Dependent variableRETt,t+3 Dependent variableRETt,t+12
Intercept 0.100 (0.01) 0.107 (0.01) 0.400 (0.01) 0.386
(0.01)SIOt 0.077 (0.01) 0.189 (0.01)LIOt 0.007 (0.63) 0.001
(0.99)SIOt1 0.049 (0.01) 0.157 (0.02)LIOt1 0.005 (0.67) 0.005
(0.91)SIOt 0.062 (0.02) 0.223 (0.01)LIOt 0.006 (0.72) 0.037
(0.32)BM 0.005 (0.09) 0.005 (0.07) 0.021 (0.01) 0.021 (0.03)MKTCAP
0.007 (0.01) 0.008 (0.01) 0.029 (0.01) 0.028 (0.01)VOL 0.062 (0.47)
0.081 (0.37) 0.306 (0.35) 0.269 (0.41)TURN 0.185 (0.01) 0.186
(0.01) 0.689 (0.01) 0.694 (0.01)PRC 0.002 (0.61) 0.001 (0.77) 0.022
(0.23) 0.023 (0.22)SP500 0.017 (0.01) 0.017 (0.01) 0.059 (0.01)
0.058 (0.01)RET3,0 0.003 (0.84) 0.001 (0.97) 0.121 (0.01) 0.120
(0.01)RET12,3 0.028 (0.01) 0.028 (0.01) 0.060 (0.01) 0.060
(0.01)AGE 0.003 (0.06) 0.003 (0.05) 0.008 (0.17) 0.008 (0.18)DP
0.018 (0.70) 0.025 (0.61) 0.123 (0.49) 0.142 (0.45)Avg. R2 0.082
0.084 0.089 0.090
918
The Review of Financial Studies / v 22 n 2 2009
-
Institutional Investors and Equity Returns
Table 9(Continued)
Panel B: 19922003
Dependent variableRETt,t+3 Dependent variableRETt,t+12
Intercept 0.129 (0.01) 0.130 (0.01) 0.555 (0.01) 0.553
(0.01)SIOt 0.040 (0.02) 0.126 (0.05)LIOt 0.016 (0.29) 0.046
(0.37)SIOt1 0.036 (0.03) 0.114 (0.10)LIOt1 0.027 (0.09) 0.067
(0.15)SIOt 0.062 (0.01) 0.162 (0.08)LIOt 0.047 (0.03) 0.047
(0.46)BM 0.009 (0.01) 0.010 (0.01) 0.048 (0.01) 0.049 (0.01)MKTCAP
0.005 (0.01) 0.005 (0.01) 0.019 (0.01) 0.018 (0.01)VOL 0.009 (0.92)
0.006 (0.95) 0.035 (0.92) 0.037 (0.91)TURN 0.051 (0.15) 0.046
(0.20) 0.204 (0.03) 0.193 (0.05)PRC 0.018 (0.01) 0.018 (0.01) 0.060
(0.01) 0.060 (0..01)SP500 0.018 (0.01) 0.017 (0.01) 0.062 (0.01)
0.060 (0.01)RET3,0 0.004 (0.73) 0.004 (0.73) 0.099 (0.01) 0.097
(0.01)RET12,3 0.015 (0.01) 0.015 (0.01) 0.007 (0.60) 0.007
(0.59)AGE 0.002 (0.36) 0.001 (0.41) 0.000 (0.99) 0.001 (0.91)DP
0.060 (0.03) 0.067 (0.02) 0.350 (0.01) 0.365 (0.01)Avg. R2 0.080
0.081 0.068 0.070
This table summarizes the results of cross-sectional regressions
of one-quarter-ahead orone-year-ahead returns on short-term and
long-term institutional ownership, and otherstock characteristics.
The sample period is from 1980 to 1991 in Panel A and is from1992
to 2003 in Panel B. Institutional holdings are obtained from
Thomson Financial.Stock characteristics are from the CRSP and
COMPUSTAT database. RETt,t+3 isone-quarter-ahead stock return.
RETt,t+12 is one-year-ahead stock return. IO is totalinstitutional
ownership. SIO is short-term institutional ownership. LIO is
long-terminstitutional ownership. An institutional investor is
classified as a short-term investorif its past 4-quarter turnover
rate ranks in the top tertile. An institutional investor
isclassified as a long-term investor if its past 4-quarter turnover
rate ranks in the bottomtertile. MKTCAP is market capitalization.
Age is firm age measured as number of monthssince first return
appears in the CRSP database. DP is dividend yield. DP is
winsorizedat the 99th percentile. BM is book-to-market ratio. BM is
winsorized at the 1st percentileand 99th percentile. PRC is share
price. TURN is the average monthly turnover overthe previous
quarter. VOL is the monthly volatility over the past two years.
SP500 isdummy variable for S&P 500 index membership. RETt3,t is
the lagged three-monthreturn. RETt12,t3 is the lagged nine-month
return preceding the beginning of thequarter. All variables except
institutional ownership, S&P 500 index membership, andstock
returns are expressed in natural logarithms. We estimate
cross-sectional regressionsfor each quarter. We use the Fama and
MacBeth (1973) methodology and report thetime-series average
regression coefficients. Numbers in parentheses are p-values
basedon NeweyWest standard errors. We use the Fama and MacBeth
(1973) methodology.Regression coefficients on institutional
ownership and trading that are statisticallysignificant at the 5%
levels are in bold. All returns are in percent.
3.7 Are our results on long-term institutions driven by index
funds?Our results indicate that long-term institutional holdings or
trading haveno explanatory power for future stock returns. One
possibility is thatsome long-term institutions in our sample are
institutions specializing inindex funds. Since the primary
objective of index funds is to track theperformance of their
respective indexes, they have little incentive to
collectinformation about future returns. To examine if index funds
drive ourresults regarding long-term institutions, we conduct two
tests.
919
-
In the first test, we split the sample period into two halves.
During thefirst half of our sample period, index funds are
relatively undeveloped.The average share of index funds among all
equity funds is only about1% from 1980 to 1991.9 If long-term
institutions can predict future stockreturns but their
predictability is diluted by index funds, we would expectlong-term
institutions to have stronger predictive power during the firsthalf
of our sample period. Panel A of Table 9 reports the results forthe
period from 1980 to 1991, and Panel B reports the results for
theperiod from 1992 to 2003. Our sub-period results are largely
similar tothose for the entire sample period. Even for the first
half of our sampleperiod, neither the holdings nor trading by
long-term institutions predictfuture stock returns. This is true
for both one-quarter-ahead returns andone-year-ahead returns. These
results suggest that index funds do notdrive our results regarding
long-term institutions. We also note that short-term institutional
holdings and trading strongly predict future returns inboth
sub-periods. This shows that our main results regarding
short-terminstitutions are robust to alternative sample
periods.
In the second test, we exclude from our sample those fund
families thateither have more than $1 billion in equity index funds
or more than 50%of their total net assets in equity index funds on
average from 1992 to2002.10 We then re-estimate regression (7). The
results are reported inTable 10. After excluding families
specializing in index funds, we still findno evidence that the
holdings or trading by long-term institutions predictfuture stock
returns. In summary, these additional tests suggest that ourresults
on long-term institutions do not seem to be driven by index
funds.
4. Conclusions
This article finds a significant relation between institutions
investmenthorizons and their informational roles. We show that the
positive relationbetween institutional ownership and future stock
returns documented inGompers and Metrick (2001) is driven by
short-term institutional investors.More importantly, we find strong
evidence that changes in short-terminstitutional ownership also
predict future returns. By contrast, we findno evidence that
long-term institutions holdings or trading predict futurereturns.
These results are consistent with the hypothesis that
short-terminstitutions are better informed.
If short-term institutions have an informational advantage,
theiradvantage should be greater for firms with smaller size and
more growth
9 We calculate the share of index funds using data from the CRSP
survivor-bias free mutual fund database.We